WO2015120243A1

WO2015120243A1 - Application execution control utilizing ensemble machine learning for discernment

Info

Publication number: WO2015120243A1
Application number: PCT/US2015/014769
Authority: WO
Inventors: Ryan PERMEH; Derek A. SOEDER; Glenn Chisholm; Braden RUSSELL; Gary Golomb; Matthew Wolff; Carl A. KUKKONEN
Original assignee: Cylance Inc.
Priority date: 2014-02-07
Filing date: 2015-02-06
Publication date: 2015-08-13
Also published as: US20150227741A1; JP2017508210A; EP3103070B1; US10235518B2; CA2938580C; US20190188375A1; HK1232326A1; JP6662782B2; CA2938580A1; AU2015213797B2; EP3103070A1; AU2015213797A1; US10817599B2; WO2015120243A8

Abstract

Described are techniques to enable computers to efficiently determine if they should run a program based on an immediate (i.e., real-time, etc.) analysis of the program. Such an approach leverages highly trained ensemble machine learning algorithms to create a real-time discernment on a combination of static and dynamic features collected from the program, the computer's current environment, and external factors. Related apparatus, systems, techniques and articles are also described.

Description

Application Execution Control Utilizing Ensemble Machine

Learning For Discernment

RELATED APPLICATION

[0001] This application claims priority to U.S. Pat. App. Ser. No. 61/937,379 filed on February 7, 2014, the contents of which are hereby fully incorporated by reference.

TECHNICAL FIELD

[0002] The subject matter described herein relates to techniques for selectively allowing applications to execute that utilize ensemble machine learning models.

BACKGROUND

[0003] Conventional techniques of application execution control for programs run on computer systems rely on static methods such as databases of signatures to determine if a computer can safely run a particular program. Existing application control systems require frequent updates to these databases, and require significant overhead to manage this process. Additionally, their ability to control execution efficiently and correctly reduces as their databases grow. Such approaches utilize significant resources (e.g., memory, CPU, etc.) and additionally have a high management overhead.

SUMMARY

[0004] The current subject matter is directed to enabling computers to efficiently determine if they should run a program based on an immediate (i.e., real-time, etc.) analysis of the program. This approach leverages highly trained ensemble machine learning algorithms to create a real-time discernment on a combination of static and dynamic features collected from the program, the computer's current environment, and external factors.

[0005] In one aspect, data is received (i.e., received from a remote data source, loaded into memory, accessed from local or connected storage, etc.) that includes at least one feature associated with a program. Thereafter, it is determined, based on the received data and using at least one machine learning model, whether to allow the program to execute or continue to execute (if it is already executing). The program executes or continues to execute if it is determined that the program is allowed to execute. Otherwise, the program is prevented from executing or continuing to execute if it is determined that the program is not allowed to execute.

[0006] One or more of the utilized machine learning models can be trained using feature data derived from a plurality of different programs. In addition or in the alternative, one or more of the machine learning models can be trained using supervised learning. Further in addition or in the alternative, one or more of the machine learning models can be trained using unsupervised learning.

[0007] The at least one feature of the program can be collected by a feature collector. The feature collector can collect features at a pre-specified point in time (e.g., at commencement of execution of the program or subsequent to execution of the program).

[0008] The at least one feature collected by the feature collector can include a combination of point in time measurements and ongoing measurements during execution of the program. The at least one feature collected by the feature collector can include one or more operational features that are passively collected prior to execution of the program, and such operational features can be stored in a cache.

[0009] The at least one feature can include at least one operational feature that characterizes an operational environment of a system to execute the program. The at least one operational feature can include one or more of: program reputation, contextual information, system state, system operating statistics, time-series data, existing programs, operating system details, program run status, and configuration variables.

[0010] The at least one features can include at least one static feature that characterizes the program. The at least one static feature can be, for example, measurements of the program, structural elements of the program, or contents of the program.

[0011] The at least one feature can include at least one dynamic feature that characterizes execution of the program. The at least one dynamic feature can include, for example, interactions with an operating system, subroutine executions, process state, program or system execution statistics, or an order of an occurrence of events associated with the program.

[0012] The at least one feature can include at least one external feature from a source external to a system to execute the program. The external feature or features can be obtained, for example, from at least one remote database or other data source.

[0013] At least one feature can take a format selected from a group consisting of: binary, continuous, and categorical. [0014] The at least one machine learning model can include an ensemble of machine learning models. The ensemble of machine learning models can include one or more models such as neural network models, support vector machine models, scorecard models, logistic regression models, Bayesian models, decision tree models or other applicable classification models. An output of two or more machine learning models can be combined and used to determine whether or not to allow the program to execute or continue to execute.

[0015] The determination can include generating a score characterizing a level of safety for executing the program. The generated score can be used to determine whether or not to allow the program to execute. The determination can also include generating a confidence level for the generated score that is used to determine whether or not to allow the program to execute.

[0016] Preventing the program from executing or continuing to execute can include at least one of many actions. These actions can include one or more of: blocking at least a portion of the program from loading into memory, determining that a dynamic library associated with the program is unsafe, blocking the dynamic library associated with the program from loading into memory, unloading a previously loaded module (portion of code, etc.) associated with the program, disabling the program while it is running, implementing constraints on the program prior to it being run or before it continues to run, quarantining at least a portion of the program, or deleting at least a portion of the program.

[0017] In some cases, preventing the program from executing or continuing to execute can include one or more of preventing the program from executing individual operations, by modifying an access level of the program, selectively blocking attempted operations, or preventing an attempted operation and instead causing an alternative operation.

[0018] Non-transitory computer program products (i.e., physically embodied computer program products) are also described that store instructions, which when executed by one or more data processors of one or more computing systems, cause at least one data processor to perform operations herein. Similarly, computer systems are also described that may include one or more data processors and memory coupled to the one or more data processors. The memory may temporarily or permanently store instructions that cause at least one processor to perform one or more of the operations described herein. In addition, methods can be implemented by one or more data processors either within a single computing system or distributed among two or more computing systems. Such computing systems can be connected and can exchange data and/or commands or other instructions or the like via one or more connections, including but not limited to a connection over a network (e.g. the Internet, a wireless wide area network, a local area network, a wide area network, a wired network, or the like), via a direct connection between one or more of the multiple computing systems, etc.

[0019] The subject matter described herein provides many advantages. For example, the current subject matter provides more rapid discernment while, at the same time, consuming fewer resources such as memory and processors.

[0020] The details of one or more variations of the subject matter described herein are set forth in the accompanying drawings and the description below. Other features and advantages of the subject matter described herein will be apparent from the description and drawings, and from the claims.

DESCRIPTION OF DRAWINGS

[0021] FIG. 1 is a system diagram illustrating elements used to provide application execution control discernment;

[0022] FIG. 2 is a diagram characterizing ensemble discernment; and [0023] FIG. 3 is a process flow diagram illustrating a method for discernment using at least one machine learning model.

DETAILED DESCRIPTION

[0024] The current subject matter can be implemented, in some examples, using three major elements to produce an efficient method of discernment. In this regard, discernment refers to the characterization of whether or not to allow a particular application / application module to execute on a particular computing system or systems.

These major software elements are illustrated in diagram 100 of FIG. 1 and can include a feature collection system 110 (sometimes referred to as a feature collector), a discernment engine 120, and an enforcement system 130. The feature collection system

110 collects or otherwise accesses features characterizing a program and/or the environment in which the program is being executed or to be executed. These features are passed on to the discernment engine 120 which can make a decision on whether or not to allow the program to execute. If it is determined that the program should not execute, the enforcement system 130 takes action to prevent the application from executing / continuing to execute. [0025] A "feature" as used herein can include any salient data / data point that can be used to measure the implied safety of a potentially run program. A "program" as used herein is a piece of executable computer code that a user or system wishes to execute, and may include associated data / metadata. "Discernment" as used herein is the process of deciding whether the program should be executed or not (including whether or not to continue executing a program). "Enforcement" as used herein is a process in which the effects of discernment are made effective in a computer system. The current subject matter can utilize one or more machine learning models that are each a mathematically based understanding of a particular situation and one or more algorithms defined to determine an outcome from a particular input against the model. In some variations, an ensemble of machine learning models can be used which is a collection of models utilized in a particular way to generally improve accuracy or reduce variance.

[0026] The current subject matter offers an effective method of application control that differs from traditional approaches in a variety of ways. Traditional approaches utilize either the concept of a "blacklist", or a set of programs to explicitly disallow, or a "whitelist", or a set of programs to explicitly allow. The current subject matter foregoes both as primary selection criteria and instead measures various features from the system and uses these against a previously trained machine learning model and/or ensemble of machine learning models.

[0027] The ensemble of machine learning models can be devised and trained before application control. Due to the predictive nature of various machine learning algorithms, a trained model allows a "fuzzy" match against safe and unsafe programs. By carefully selecting and training the models in the ensemble, the system can act resiliently against change over time, accommodating small and large changes in program behaviors that resemble "safety" or a lack thereof. A machine learning model may be characterized by an algorithm it incorporates, which may include, as an example, neural networks, support vector machines, logistic regressions, scorecard models, Bayesian algorithms, and decision trees. A machine learning model can be trained using supervised learning, in which a training set of input samples labeled with the desired output values conditions the model to correctly classify samples that do not occur in the training set, or it may be trained using unsupervised learning, in which an algorithm identifies hidden structure in unlabeled data. Reinforcement learning represents a third process for training a model.

[0028] Referring back again to diagram 1 of FIG. 1, the feature collector 110 can send passive features (operational and dynamic) on an ongoing basis to the discernment engine 120. The discernment engine 120 can request point in time features from the feature collector 110 at a particular decision point, such as execution. These point in time features can include observations about the computer's state extrinsic to the program or related features from an external source. The discernment engine 120 can then decide if the program should execute. If execution is allowed, the program executes; if execution is disallowed, the enforcement system 130 prevents the application from executing.

[0029] FIG. 2 is a diagram 200 characterizing ensemble discernment in which an original vector 210 can be passed to the discernment engine 120 for scoring 230. The discernment engine 120 can use a model selector 220 to choose one or more models to run (in this example, Models A, B, C). The selection of a model can be be predicated on features provided by the feature collector 110, a user configuration, the current availability or scarcity of computing resources, and/or other state information. Each such model can be comprised of several possible algorithms. The output of the various algorithms and models can be combined (using, for example, a weighting arrangement or model) in a scoring component 230. A final output can be a decision (or in some cases a score) characterizing the results and a confidence level.

[0030] Feature collection can be a combination of point in time and ongoing measurements, and can include the passive collection of features into a general cache. Features can be used to generate data points for which the discernment engine 120 makes a decision. The discernment engine 120 can utilize the features collected to make a decision based on previously collected data. The enforcement system 130 can implement the technical details of operation regarding the decisions made from the discernment engine 120.

[0031] If a user or other program wishes to execute a program, it will first ask the discernment engine 120 to decide if this is a positive action. The discernment engine 120 can either answer with previous discernments, or create a new discernment using a combination of previously collected features and features collected via a point in time analysis. With the decision made, the enforcement system 130 can implement the logic to allow or disallow execution of the program, and any other elements necessary to implement the discernment decision in an ongoing manner.

[0032] Features can be collected from various sources. In one

implementation, features can be collected from four primary sources. [0033] A first source can comprise operational features that relate to the operational environment of the system. Operational features can include existing programs, details about the operating system, run status of the program, configuration variables associated with the program, and other measures particular to the environment in which the program is intended to run. Some of these features can be ongoing (i.e., they are active features); others can be determined at a particular point in time (i.e., they are passive features).

[0034] A second source can comprise static features that concern the program that wishes to run. Measurements about the program itself, including structural elements and program contents, can be collected. These features can be calculated by examining the contents of the file and processing through analytic methods. One example of a static feature of a program is the size of such program. Examples of structural elements of a program can include the number of sections it comprises, the proportion of the program described by each section, and the proportion of the program not described by any section. The computed Shannon entropy of each section is an example of a feature derived from processing.

[0035] A third source can comprise dynamic features that relate to individual program execution. Dynamic features can generally be collected in an ongoing manner. The dynamic features can be associated with a particular program, rather than the system itself. These features can be used to determine potentially hostile activities from a program that was either unable to receive a high confidence discernment prior to execution or otherwise authorized to run under direct management policy. [0036] A fourth source can comprise external features that can be generally extracted from sources of information outside of the host computer itself, generally via a remote data source such as a lookup on the network. This lookup can include a query against a cloud database, or a deeper analysis of certain elements on a network based computer. For example, external features can include a determination by a trusted third party as to a program's authenticity, a program's prevalence among a larger population of computers, and/or the reputations of other computers contacted by a program.

Frequently, these features entail knowledge that is impractical to host on an individual computer due to size, complexity, or frequency of updates. Due to the latency of a network lookup, these features can generally be collected in response to a particular request from the discernment engine 120, at a particular point in time.

[0037] Features can be collected into efficient computer data structures, such as hash tables, binary trees, and vectors, and the features can be passed to the discernment engine 120. Ongoing features can be collected and held for an appropriate amount of time to ensure their ability to usefully affect the discernment process. Point in time features can be collected in an on-demand manner, typically on the event of discernment.

[0038] Features can be binary, continuous, or categorical in nature. Binary features can only be in one of two states. Continuous features can represent a value along a range, and are generally numeric in nature. Categorical features can represent a value within a discrete set of possible values.

[0039] Features can be considered first order or second order or nth order. First order features are features measured directly from the source. These features can be combined or further analyzed by various methods to generate second order features. Such further analyzing can include making a mathematical analysis of the value of a first order feature, or by applying combinations of first order features to develop a truly unique second order feature.

[0040] The discernment engine 120 can create a decision on the anticipated safety of an application. The discernment engine 120 can receive input from the feature collector 110 and apply an ensemble of machine learning models to calculate a score that determines if an application is safe to run or not, as well as a confidence in the accuracy of the score.

[0041] The discernment engine 120 can take features in combination or singly and can, in some cases, use a process known as vectorization to turn individual features into a mathematical vector. This process can involve creating a compact and efficient representation of the input. The vector can be used by the various machine learning algorithms to generate a score.

[0042] The use of ensembles allows multiple, distinct models to be tailored to suit more specialized combinations of features within the more common types of programs. Each sample can be approached with a model that is more appropriate for its type. In addition to model specificity, the general ensemble can offer multiple different learning algorithms per model. This allows sample discernment to benefit from multiple different assessments. Some specific models have lower error rates for particular algorithms, and combining them in a weighted manner helps achieve the highest results.

[0043] Ensemble models and/or their outputs can be combined using individualized measured error rates in a weighting scheme (such as a scorecard model). Each model that scores can be normalized and adjusted by its measured error rate. This final combination allows for the most accurate understanding from a variety of sources.

[0044] The enforcement system 130 can be a component that implements methods for disabling execution of a program. The enforcement system 130 can use a variety of tactics to disable execution in a safe and reliable way.

[0045] Decisions regarding a program may not always be determined before program execution, and so there may be some more complex scenarios that require additional handling. The enforcement system 130 can be integrated deeply with the computer operating system and act on behalf of the discernment engine 120.

[0046] The enforcement system 130 can implement one or more of blocking a process or dynamic library from loading into memory, unloading a previously loaded module, disabling a running program, implementing constraints on a program to be run, quarantining hostile applications, and/or deleting hostile applications. It is often desirable for the enforcement system 130 to issue an alert when a module determined to be hostile is accessed and/or when action is attempted against a hostile module.

[0047] The enforcement system 130 can utilize processes implemented both in the operating system core, and implanted in each process. These processes can allow for high degrees of control from both the core operating system level, as well as deep introspection and control from within the application itself.

[0048] Additionally, the enforcement system 130 can utilize tactics for preventing an application from running or restricting its level of access. Such tactics can include moving, renaming, or deleting the program; applying attributes or access controls to the program; forcing the application to run with reduced privileges; forcing the application to run in a "sandbox," where certain actions are redirected to access a virtualized system state; and/or other monitoring and controlling the actions an application may perform.

[0049] The systems / technique herein can go into effect when an attempt is made to run a program, or a decision is otherwise warranted by user defined behavior, such as intentionally scanning a file to ascertain its safety.

[0050] With reference again to diagram 100 of FIG. 1, the features originating from the operating system and the dynamic feature collection system 110 can continue to stream into the discernment engine 120 in an ongoing manner. These can be generally available for use within the discernment engine 120, and may initiate a discernment action if one is warranted.

[0051] Generally, however, the system / methods can be activated during the actions of the system or the user when they choose to either start an application or otherwise choose to determine a file's safety. When one of these events is triggered, the discernment engine 120 can request additional details from the feature collector. The feature collector 110 can then gather the appropriate details and pass them to the discernment engine 120. These features may originate via static, dynamic, operational, or external features.

[0052] The discernment engine 120 can take all collected features, and use a vectorization process to develop a vector as input (see diagram 200 of FIG. 2). The input vector 210 can be associated with one or more models by the model selector 220 of the discernment engine 120. For each model the model selector 220 chooses, the input vector 210 can be applied. Each model can have one or more algorithms associated with it, generating a series of individual scores. The outputs of the individual models can be combined in a scoring component 230, utilizing a weighting scheme (e.g., a scorecard model). The scoring component 230 can generate a final score, comprised of a result (e.g., safe or not) and a confidence in that result.

[0053] FIG. 3 is a process flow diagram 300 in which, at 310, data is received (i.e., accessed, obtained, etc.) that comprises at least one feature associated with a program. Thereafter, at 320, it can be determined, based on the received data and using at least one machine learning model, whether to allow at least a portion of the program to execute. The at least one machine learning model used in this regard can be trained using, for example, supervised learning and/or unsupervised learning (in some cases there may be a combination of models that use each type of learning). Subsequently, at 330, the program can execute if it is determined that at least a portion of the program is allowed to execute. Otherwise, at 330, at least a portion of the program is prevented from executing / continuing to execute if it is determined that the program (or portion thereof) is not allowed to execute.

[0054] One or more aspects or features of the subject matter described herein may be realized in digital electronic circuitry, integrated circuitry, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various implementations may include

implementation in one or more computer programs that are executable and/or

interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device (e.g., mouse, touch screen, etc.), and at least one output device.

[0055] These computer programs, which can also be referred to as programs, software, software applications, applications, components, or code, include machine instructions for a programmable processor, and can be implemented in a high-level procedural language, an object-oriented programming language, a functional

programming language, a logical programming language, and/or in assembly/machine language. As used herein, the term "machine-readable medium" (sometimes referred to as a computer program product) refers to physically embodied apparatus and/or device, such as for example magnetic disks, optical discs, memory, and Programmable Logic Devices (PLDs), used to provide machine instructions and/or data to a programmable data processor, including a machine-readable medium that receives machine instructions as a machine -readable signal. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable data processor. The machine-readable medium can store such machine instructions non-transitorily, such as for example as would a non-transient solid state memory or a magnetic hard drive or any equivalent storage medium. The machine-readable medium can alternatively or additionally store such machine instructions in a transient manner, such as for example as would a processor cache or other random access memory associated with one or more physical processor cores.

[0056] The subject matter described herein may be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front- end component (e.g., a client computer having a graphical user interface or a Web browser through which a user may interact with an implementation of the subject matter described herein), or any combination of such back-end, middleware, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), and the Internet.

[0057] The computing system may include clients and servers. A client and server are generally remote from each other and typically interact through a

communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

[0058] The subject matter described herein can be embodied in systems, apparatus, methods, and/or articles depending on the desired configuration. The implementations set forth in the foregoing description do not represent all

implementations consistent with the subject matter described herein. Instead, they are merely some examples consistent with aspects related to the described subject matter. Although a few variations have been described in detail above, other modifications or additions are possible. In particular, further features and/or variations can be provided in addition to those set forth herein. For example, the implementations described above can be directed to various combinations and subcombinations of the disclosed features and/or combinations and subcombinations of several further features disclosed above. In addition, the logic flow(s) depicted in the accompanying figures and/or described herein do not necessarily require the particular order shown, or sequential order, to achieve desirable results. Other implementations may be within the scope of the following claims.

Claims

WHAT IS CLAIMED IS:

1. A method comprising :

receiving data comprising at least one feature associated with a program;

determining, based on the received data and using at least one machine learning model, whether to allow the program to execute or continue to execute;

allowing at least a portion of the program to execute or continue executing if it is determined that the program is allowed to execute; and

preventing at least a portion of the program from executing or continuing to execute if it is determined that the program is not allowed to execute.

2. A method as in claim 1 , wherein one or more of the at least one machine learning model is trained using feature data derived from a plurality of different programs.

3. A method as in claim 1 or 2, wherein one or more of the at least one machine learning model is trained using supervised learning.

4. A method as in any of the preceding claims, wherein one or more of the at least one machine learning model is trained using unsupervised learning.

5. A method as in any of the preceding claims further comprising:

collecting the at least one feature of the program by a feature collector.

6. A method as in claim 5, wherein the feature collector collects features at a pre- specified point in time.

7. A method as in claim 6, wherein the pre-specified point in time is at execution of the program.

8. A method as in any of claim 5 to 7, wherein the at least one feature comprises a combination of point in time measurements and ongoing measurements during execution of the program.

9. A method as in any of claim 5 to 8, wherein the at least one feature comprises one or more operational features that are passively collected prior to execution of the program and the method further comprises:

storing the operational features in a cache.

10. A method as in any of the preceding claims, wherein the at least one feature comprises at least one operational feature that characterizes an operational environment of a system to execute the program.

11. A method as in claim 10, wherein the at least one operational feature is selected from a group consisting of: program reputation, contextual information, system state, system operating statistics, time-series data, existing programs, operating system details, program run status, or configuration variables.

12. A method as in any of the preceding claims, wherein the at least one features comprises at least one static feature that characterizes the program.

13. A method as in claim 12, wherein the at least one static feature is selected from a group consisting of: measurements of the program, structural elements of the program, or contents of the program.

14. A method as in any of the preceding claims, wherein the at least one feature comprises at least one dynamic feature that characterizes execution of the program.

15. A method as in claim 11, wherein the at least one dynamic feature is selected from a group consisting of: interactions with an operating system, subroutine executions, process state, program or system execution statistics, or an order of an occurrence of events associated with the program.

16. A method as in any of the preceding claims, wherein the at least one feature comprises at least one external feature from a source external to a system to execute the program.

17. A method as in claim 16, wherein the at least one external feature is obtained from at least one remote database or other data source.

18. A method as in any of the preceding claims, wherein the at least one feature takes a format selected from a group consisting of: binary, continuous, and categorical.

19. A method as in any of the preceding claims, wherein the at least one machine learning model comprises an ensemble of machine learning models.

20. A method as in claim 19, wherein the ensemble of machine learning models is selected from a group consisting of: neural network models, support vector machine models, scorecard models, logistic regression models, Bayesian models, or decision tree models.

21. A method as in claim 19, wherein an output of two or more machine learning models is combined and used to determine whether or not to allow the program to execute or continue to execute.

22. A method as in any of the preceding claims, wherein the determining comprises generating a score characterizing a level of safety for executing the program, wherein the generated score is used to determine whether or not to allow the at least a portion of the program to execute.

23. A method as in claim 22, wherein the determining further comprises generating a confidence level for the generated score, wherein the generated confidence level is used to determine whether or not to allow the at least a portion of the program to execute.

24. A method as in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises blocking at least a portion of the program from loading into memory.

25. A method as in in any of the preceding claims further comprising:

determining that a dynamic library associated with the program is unsafe;

wherein preventing the at least a portion of the program from executing or continuing to execute comprises: blocking the dynamic library associated with the program from loading into memory.

26. A method as in in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: unloading a previously loaded module associated with the program.

27. A method in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: disabling the program while it is running.

28. A method in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: implementing constraints on the program prior to it being run or before it continues to run.

29. A method as in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: quarantining at least a portion of the program.

30. A method as in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: deleting at least a portion of the program.

31. A method as in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: preventing the program from executing individual operations.

32. A method as in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: modifying an access level of the program.

33. A method as in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: selectively blocking attempted operations.

34. A method as in any of the preceding claims, wherein preventing the at least a portion of the program from executing or continuing to execute comprises: preventing an attempted operation and instead causing an alternative operation.

35. A non-transitory computer program product storing instructions which, when executed by at least one hardware data processor forming part of at least one computing device, result in a method as in any of the preceding claims.

36. A system comprising :

at least one data processor; and

memory storing instructions which, when executed by the at least one data processor, result in a method as in any of claims 1 to 34.